The transfer of risk taking along the supply chain

Abstract

We show that suppliers’ risk taking is positively influenced by that of their major customers. This result is consistent with the notion that when major customers take more risk to enhance their bargaining power and rent extraction ability, suppliers may respond by also engaging in more risk taking to improve their bargaining positions. Further cross-sectional analysis shows that the transfer of risk taking along the supply chain becomes stronger when suppliers and customers have more comparable bargaining power and when the former have greater risk-taking capacities. Our findings are robust to a series of tests addressing endogeneity concerns.

Appendices

Appendix 1: Variable definitions

Variables	Definitions
Risk-taking measures
σRET	The volatility of daily stock returns, which is computed by taking the common logarithm of the standard deviation of daily stock returns (RET) for each fiscal year. Note that we use the same formula for suppliers and their major customers and use prefixes “S_” and “C_” before each variable to indicate that it relates to suppliers and customers, respectively
σROA	The volatility of returns on assets (ROA), which is computed by taking the common logarithm of the standard deviation of ROA from year t−5 to year t + 5 where ROA is income before extraordinary items scaled by lagged total assets. We require at least 7 non-missing ROA to calculate the standard deviation
σCF	The volatility of cash flows (CF), which is computed by taking the common logarithm of the standard deviation of CF from year t−5 to year t + 5, where CF is operating income before depreciation and amortization minus interest expense, tax expense and the dividend paid to common shareholders, all scaled by total assets. We require at least 7 non-missing CF to calculate the standard deviation
wσRET	The weighted-average volatility of stock return of all major customers, which is equal to the sum of σRET times the ratio of a supplier’s sales to a major customer to its total sales, divided by the supplier’s total major customers in a given year
wσROA	The weighted-average returns on assets of all major customers, which is equal to the sum of σROA times the ratio of a supplier’s sales to a major customer to its total sales, divided by the supplier’s total major customers in a given year
wσCF	The weighted-average cash flows from operations of all major customers, which is equal to the sum of σCF times the ratio of a supplier’s sales to a major customer to its total sales, divided by the supplier’s total major customers in a given year
CAPEX	Capital expenditure, which is computed by taking the ratio of capital expenditures to lagged gross property, plant and equipment. Note that we do not use net property, plant and equipment in the computation of CAPEX as we want to avoid the effect of managerial discretion in estimating depreciation (i.e., to avoid the effect of earnings management) (Teoh et al. 1998)
INTCAP	This is intangible capital. Following Peters and Taylor (2017), we compute this variable by taking the sum of firms’ externally purchased intangibles ($X$) and internally generated intangibles ($I$). X is intangible assets (item intan, where missing values are replaced by 0). $I$ is the sum of knowledge capital ($G$) and organisational capital ($O$). Knowledge capital is calculated as: ${G}_{it}=\left(1-{\delta }_{R\&D}\right){G}_{it-1}+{R\&D}_{it}$, where ${G}_{it}$ is knowledge capital of firm i at the end of year t. ${\delta }_{R\&D}$ is the $R\&D$ depreciation rate based on Li and Hall (2019, Table 2)’s table of industry depreciation (Table 2). When firms’ industry is not in this table, ${\delta }_{R\&D}$ is set to be 15% (Peters and Taylor 2017; Li and Hall 2019). Meanwhile, organisational capital is as: ${O}_{it}=\left(1-{\delta }_{SG\&A}\right)*30\%*{SG\&A}_{it-1}+30\%*{SG\&A}_{it},$ where ${O}_{it}$ is organisational capital of firm i at the end of year t. $SG\&A$ is selling and administration expense (item xsga). ${\delta }_{GS\&A}$ is the depreciation rate of $SG\&A$ spending as an investment in intangible capital. Following Peters and Taylor (2017), we treat 30% of $SG\&A$ as investment in organisational capital and ${\delta }_{GS\&A}$ is set at 20%
Firm characteristics
Lev	Financial leverage, which is the ratio of the sum of long-term and short-term debts to total assets
Liquidity	The ratio of cash and short-term investments to total assets
LnAge	The common logarithm of firm age, which is the number of years firm i has appeared in the Compustat database (i.e., the current fiscal year minus the first year the firm appears in the database)
Loss	A dummy variable which takes a value of 1 if firms’ ROA is negative and 0 otherwise
R&D	Investment in research and development (R&D) activities, which is equal to R&D expenses scaled by lagged total assets. Missing values are replaced by 0
Size	Firm size, which is equal to the common logarithm of total assets
Distress	A dummy variable for financial distress, which is equal to 1 if firms have an Altman (1968)’s Z-SCORE lower than 1.81 (i.e., considered to be in the “distress” zone), and 0 otherwise
Rating	S&P Domestic Long-Term Issuer Credit Rating (in the Compustat database). Following Tewari et al. (2015), we convert S&P ratings to ordinal ratings, where ordinal ratings are assigned the highest value of 16 if rated as AAA (prime issues) and the lowest value of 1 if rated as B- (highly speculative issues). Ratings lower than B- and unrated cases are assigned a value of 0
Patent	The patents variable (Kogan et al. 2017), which is the common logarithm of the sum of one plus the number of patents filed and eventually granted in a given year. Patent and citation data are provided by Professor Noah Stoffman and available at https://kelley.iu.edu/nstoffma/
Citation	Citations on patents (Kogan et al. 2017), which is the common logarithm of the sum of one plus truncation-adjusted citations received on all patents filed by a firm in a given year, where the truncation-adjusted citations on each patent are equal to the number of citations adjusted for truncation using the index of Hall et al. (2001). Patent and citation data are provided by Professor Noah Stoffman and available at https://kelley.iu.edu/nstoffma/
Ncskew	A measure of stock price crash risk, which is firm-specific negative skewness of weekly returns (Chen et al. 2001; Kim et al. 2011). Weekly returns are calculated as log(1 + ${\varepsilon }_{j,\tau }$), where ${\varepsilon }_{j,\tau }$ is the residual of the following regression: ${r}_{j,\tau }={\beta }_{0}+{\beta }_{1,j}{r}_{m,\tau -1}+{\beta }_{2,j}{r}_{i,\tau -1}+{\beta }_{3,j}{r}_{m,\tau }+{\beta }_{4,j}{r}_{i,\tau }++{\beta }_{5,j}{r}_{m,\tau +1}+{\beta }_{6,j}{r}_{i,\tau +1}+{\varepsilon }_{j,\tau }$, where ${r}_{j,\tau }$ is the stock return of firm j in week $\tau$; ${r}_{m,\tau }$ and ${r}_{i,\tau }$ are the CRSP value-weighted market index and the Fama–French value-weighted industry index in week $\tau$, respectively
Duvol	A measure of stock price crash risk, which is the common logarithm of the ratio of the standard deviation of firm-specific down-week to up-week weekly returns
Dividend	Dividend paid to common shareholders scaled by lagged total assets
CF	A measure of operating cash flow which is computed as operating profit before depreciation minus interest expense, tax expense and dividend paid to common shareholders, all scaled by lagged total assets
Sales_Growth	Change in sales scaled by lagged total assets
MB	The market to book ratio, which is equal to market values plus total assets (AT) minus book value of equity (CEQ) and deferred tax (TXDB), all scaled by lagged total assets
Credit	Trade credit, which is equal to trade receivables scaled by lagged total assets
Segment	The number of operating segments reported in the Compustat Segments Historical with valid two-digit SIC codes
Customer	The number of customers reported in the Compustat Segments Historical
CC	CC is the level of customer concentration, which is captured by the Herfindahl–Hirschman Index (Lian 2017). It is equal to the sum of squares of the share of sales to each customer, scaled by the firm’s total sales. It is computed as follows: $CC=\sum_{j=1}^{n}{\left(\frac{{Sales \,to \,major \,customer \,firm}_{i,j,t}}{{Total \,sales}_{i,t}}\right)}^{2}$ In this equation n is the number of corporate customers of firm i in year t reported in the Compustat Segment database and j is the major customer firm j. We keep only customer firms that belong to the customer category “COMPANY”. Also, in computing the CC for the sample firms, we keep all the customer firms with missing identifiers since it does not require such identifiers. By definition, CC ranges from 0 to 1; for a given supplier firm, it takes the value of 1 when the firm has only one customer
Measures of customer–supplier relationships and risk-taking capacities
MULTICUS	A dummy variable equal to one if a supplier has more than one major customers, and zero otherwise
C_Highrisk	An indicator for customers having high risk, which is equal to 1 if the major customer has σRET greater than or equal to the median level of all firms and 0 otherwise
DIG	A dummy variable equal to 1 if a supplier is classified, based on its two-digit SIC code, as a differentiated goods manufacturer and 0 if classified as a standardized goods manufacturer. Differentiated and standardized goods are classified based on the industry classification, following Rauch (1999) and Giannetti et al. (2011, p. 1296). Specifically, a firm is considered as a differentiated goods producer if its two-digit SIC code is 25, 27, 30, 32, 34, 35, 36, 37, 38, or 39; and as a standardized goods producer if its two-digit SIC code is 12, 14, 20, 22, 23, 24, 26, 28, 29, 31, or 35
DPRO	A dummy variable that takes a value of 1 if the supplier is a durable product manufacturer and 0 if it is a non-durable product manufacturer. Similarly to Lian (2017), we classify a supplier as a durable product manufacturer if it has a SIC code between 3400 and 3990 and as a non-durable product manufacturer if it has a SIC code between 2000 and 3390
FS	A dummy variable which is equal to 1 if the supplier’s firm size is greater than or equal to the median level of all suppliers in the sample and 0 otherwise
RATED	A dummy variable equal to 1 if the supplier has non-missing S&P Domestic Long-Term Issuer Credit Rating and 0 otherwise

Appendix 2: The matching procedure

We follow previous studies (Fee and Thomas 2004; Lian 2017; Chu et al. 2019) and match suppliers and their major customers using the Compustat Segment Customer file. The matching steps are as follows. First, we start with all US public firms in the CRSP/Compustat database for the period from 1980 to 2016. We remove firms with missing total assets, financial institutions (i.e., firms with Standard Industrial Classification (SIC) code from 6000 to 6999), and heavily regulated firms (i.e., firms with a SIC code in the range 4000–4099, 4500–4599, 4800–4899, or 4900–4999). Meanwhile, we obtain data on sales to major customers from the Compustat Segment Customer file. From these data, we delete all observations with nonsensical customer names (e.g., “1 CUSTOMER” or “NOT REPORTED”), general customer names (e.g., “ASIA PACIFIC” or “ASIA MARKET”), and missing values for sales to customers, since there is insufficient information to identify these customers and/or determine how important they are to their suppliers.

Second, we use a text-matching method to match customers’ names from the Compustat Segment Customer file with their names in the CRSP/Compustat database. We employ a matching method similar to that used in previous studies (Fee and Thomas 2004; Eshleman and Guo 2013; Gu and Venkateswaran 2017; Lian 2017; Chu et al. 2019; Nguyen et al. 2021). The text-matching method finds and matches similar sequences of letters to generate a list of potential matches to customers’ names. When a customer’s name in the Compustat Segment Customer file is the same as a firm’s name in the CRSP/Compustat database, we have an exact match. However, when a customer’s name in the Compustat Segment Customer file is just similar to a firm’s name in the CRSP/Compustat database, we manually verify the match based on SEC filings, firms’ websites, and other sources such as Bloomberg’s firm information. We classify similar matches into true matches and false matches and keep only the true matches, where we believe that the customers’ and firms’ names refer to same entities. This matching method produces true matches even when customers’ names are written as abbreviations (Fee and Thomas 2004). However, in some cases, customers’ names are written using very short abbreviations and there are therefore large differences between the disclosed customers’ and firms’ names. In these situations, following previous studies that employ the same method, we make best guesses about customers’ names and look for firms with those names in the CRSP/Compustat database. For example, the abbreviation “ASSD DRY GD” is disclosed as a major customer of APPLIED DEVICES (1981) and SELIGMAN & LATZ INC (from 1982 to 1984). In this case, we make a best guess and manually match “ASSD DRY GD” with “ASSOCIATED DRY GOODS CORP.” Following this rigorous procedure, we obtain 30,707 major customer–supplier firm-year observations, and 22,873 largest customer–supplier firm-year observations. This sample is equivalent to that of Lian (2017) and Chu et al. (2019).

In the final step, we exclude customer–supplier pairs in which customers are financial or heavily regulated firms, and those with missing stock prices or financial data needed for the main regressions. Our final sample includes 10,875 firm-year pairs of suppliers and their largest customers with sufficient data for the empirical analyses. This sample consists of 2422 unique suppliers and 791 unique customers (3439 unique pairs).